Replacing column values in a pandas DataFrame

In-depth exploration of replacements

Converting categorical data

When dealing with categorical data, such as changing 'female' to '1' and 'male' to '0', the use of map proves to be efficient:

# 'female' in code, '1' in our hearts
gender_map = {'female': 1, 'male': 0}
df['Gender'] = df['Gender'].map(gender_map)


Conditional replacement with loc

Sometimes, you want a variable to reflect certain conditions. loc combined with boolean indexing can fulfill this:

# Elevating users over 50 to 'Senior' status.
df.loc[df['Age'] > 50, 'AgeCategory'] = 'Senior'
df.loc[df['Age'] <= 50, 'AgeCategory'] = 'Adult'


Numeric conversions post-replacement

After swapping text-based labels with numbers, ensure the data type reflects these changes. This is your weapon of choice, pd.to_numeric():

# Translating 'low', 'medium', 'high' scores to '1', '2', '3'.
df['Score'] = df['Score'].replace(['low', 'medium', 'high'], [1, 2, 3])
df['Score'] = pd.to_numeric(df['Score'])


Hacks for handling replacements

Precision in indexing

Ensure you double-check column and row indices to avoid an asymmetric Matrix situation:

# Only the chosen ones (rows 2 through 4 in column 'A') will get to see 'ten'.
df.loc[2:4, 'A'] = df.loc[2:4, 'A'].replace({10: 'ten'})


Preserving NaNs during replacement

Preserving NaN values during pandas transformation? There's an app, err, replace for that:

df['A'] = df['A'].replace({10: 'ten', 20: 'twenty'}, inplace=True)


Ditching for-loops for element-wise operations

Basically, apply or vectorized operations are Indy cars, while for-loops are bicycles. For example, np.where:

import numpy as np

# It's not you, it's me(numeric 10).
df['A'] = np.where(df['A'] == 10, 'ten', df['A'])
